Sep 11 – 13, 2025
Campus Luigi Einaudi
Europe/Rome timezone

LLMind to Enhance Diagnostic Decision-Making in Mental Health: Synergy Between AI Language Models and Clinical Expertise

Sep 12, 2025, 12:00 PM
15m
Aula A1 (Pareto)

Aula A1 (Pareto)

Speaker

Andrea Spoto (Dipartimento di Psicologia Generale - Università di Padova)

Description

Background and aim: This study investigates the application of large language models (LLMs) as collaborative tools for mental health professionals, focusing on their integration with the ICD-11 framework. At the intersection of artificial and human cognition, we introduce LLMind Chat, an AI-driven interface to augment clinical reasoning by combining real-time retrieval of ICD-11 guidelines with advanced natural language processing.
Methods: Built on a Retrieval Augmented Generation (RAG) architecture using Gemma 2 (27B parameters), the system dynamically synthesizes diagnostic criteria, case histories, and evidence-based insights to support clinicians in complex decision-making scenarios. LLMind Chat’s validation framework involved comparative analysis against the DSM-5-TR Clinical Cases manual, supplemented by expert clinicians’ evaluations.
Results: Results indicate strong alignment between the model’s diagnostic suggestions and clinical judgments (76.1% accuracy with moderate inter-rater agreement) particularly in cases involving anxiety-related and psychotic disorders-areas where ICD-11’s approach benefits from computational precision. The tool prominently reduced classification errors most in disorders with overlapping symptomatology (e.g., trauma-related conditions).
Conclusions: By emphasizing the collaborative potential of AI rather than autonomous diagnosis, this work underscores how LLMs can extend clinicians’ cognitive capacities while preserving human oversight. The system’s dual chat modes -interactive case exploration and validation feedback- create a bidirectional learning loop refining algorithmic performance and clinical engagement. As mental health care navigates the complexities of modern classification systems, LLMind Chat exemplifies the transformative role of AI in bridging technological capabilities with psychological expertise, fostering a new paradigm where artificial and human intelligence coexist to enhance diagnostic accuracy and patient outcomes.

If you're submitting a symposium talk, what's the symposium title? Menti artificiali e menti umane: l'intelligenza al confine tra psicologia e tecnologia
If you're submitting a symposium, or a talk that is part of a symposium, is this a junior symposium? No

Primary author

Andrea Spoto (Dipartimento di Psicologia Generale - Università di Padova)

Co-authors

Anna Panzeri (Dipartimento di Psicologia Generale, Università degli Studi di Padova, Italia) Dr Marco Cremaschi (Department of Informatics, Systems and Communication, University of Milano Bicocca) Dr Davide Di Tolve (Department of Informatics, Systems and Communication, University of Milano Bicocca) Dr Cesare Curcio (Department of Informatics, Systems and Communication, University of Milano Bicocca) Prof. Andrea Maurino (Department of Informatics, Systems and Communication, University of Milano Bicocca)

Presentation materials

There are no materials yet.